In this study, we determined genetic loci for SCAD from five GWASs (5, 15, 16, 29, 30). Genes and top SNPs in the identified loci were collected from the GWAS papers and related supplementary online files (Figure 1). Specifically, genome-wide summary statistics were obtained from a recent GWAS (15). This GWAS comprised 270 SCAD cases and 5,263 controls of European descent. The raw data used in the present analysis were the downloaded summary results from the initial GWAS, which included association P-values of 607,778 genotyped variants with SCAD (15). The SCAD GWAS summary datasets were publicly available at the NHGRI-EBI GWAS Catalog under accession number GCST90000582. SNPs with P-values less than 1.0 × 10^–5 were selected and the SCAD loci were determined. The genome-wide significance level was set to 5.0 × 10^–8 in this study. The BP loci were identified in a GWAS examined associations between more than 10 million SNPs and systolic BP (SBP) and diastolic BP (DBP) in 317,756 individuals enrolled in the UK Biobank (17). SNPs associated with both BP and SCAD were identified.

We carried out pQTL analysis to obtain relevant evidence for the identified SNPs shared by BP and SCAD. The associations between SNPs and circulating protein levels were searched in five pQTL studies. First, the pQTL data of the KORA study which contained associations between 509,946 SNPs and 1,124 proteins were downloaded from the pGWAS Server (21).¹ The second pQTL study performed genome-wide testing of 10.6 million imputed autosomal variants against levels of 2,994 circulating proteins in 3,301 individuals of European descent from the INTERVAL study (23). The summary data are available at http://www.phpc.cam.ac.uk/ceu/proteins/. The third pQTL study analyzed 83 proteins measured in 3,394 individuals (20).² The fourth pQTL study provided the results of GWAS of 71 high-value cardiovascular disease proteins in 6,861 Framingham Heart Study participants (24). In the fifth pQTL study, 4,137 proteins covering most predicted extracellular proteins were measured in the serum of 5,457 Icelanders over 65 years of age (22). pQTL signals with P-values less than 1.0 × 10^–5 were considered in this study.

The purpose of gene ontology (GO) is to unify biological factors while integrating specific definitions, clear structures, and controlled vocabulary into a database (31). GO analysis was performed to explore the biological properties of the genes coding the identified proteins. The Database for Annotation, Visualization and Integrated Discovery (DAVID) database,³ an online bioinformatics tool, was used to illustrate the functional enrichment analysis (32). The advantages of DAVID include the detailed analysis and classification of gene and protein functions. A false positive rate less than 0.1 was considered significant.

To obtain supporting evidence for proteins identified in pQTL analysis, we performed MR analysis to examine the associations between protein levels and BP. The protein data used in these MR analyses were from the pQTL studies described above. The BP GWAS dataset comprised the summary statistics for the association between more than 10 million SNPs and SBP and DBP, which were evaluated in 317,756 individuals enrolled in the UK Biobank (17). This dataset can be downloaded at https://data.broadinstitute.org/alkesgroup/UKBB/. Summary data necessary to MR analysis were unavailable for SCAD, which kept us from examining the associations between protein levels and SCAD.

Instead, we explored the role of the identified proteins in aortic remodeling, which were highly related to BP. GWAS summary data of aortic aneurysm and dissection (AAD) (ICD-10 code: I71), which contains 452,264 individuals from the UK Biobank (30), was obtained. These GWAS included 1,470 AAD patients and 450,794 controls. The genome-wide association analysis of the UK Biobank GWAS comprised 62,394 genotyped variants and 9,113,133 imputed variants that passed quality control. The summary data for AAD can be found by searching “Aortic aneurysm and dissection” at http://geneatlas.roslin.ed.ac.uk/downloads/. In addition, summary data from GWAS of ascending aortic (AA) and descending aortic (DA) diameters were obtained. AA and DA diameters were evaluated in cardiac magnetic resonance images from the UK Biobank by a deep learning model (33).

We employed the weighted median, inverse-variance weighted (IVW) MR (34), MR-Egger (35), MR pleiotropy residual sum and outlier (MR-PRESSO) (36) and the Causal Analysis Using Summary Effect estimates (CAUSE) (37) methods to test for potential causal relationships between circulating protein levels and BP and SCAD. The inverse-variance weighted method combines the ratio estimates from each IV in a meta-analysis model (34). If the associations with circulating protein levels were to lead to horizontal pleiotropy, the intercept from MR-Egger would be expected to differ from zero (35). In such cases, we interpreted the coefficient from MR-Egger as being the more valid causal estimate. Conversely, in the absence of statistical evidence for horizontal pleiotropy from the intercept on MR-Egger, we used the IVW MR analysis as it retains greater power. The IVW MR and MR-Egger analyses were performed by using the MendelianRandomization R package (38). We also detected horizontal pleiotropy and outlier-corrected causal estimation by using MR-PRESSO tests (36). The outlier test in MR-PRESSO is the procedure to test for the MR assumption of no pleiotropy. The source code and documents for MR-PRESSO are available at https://github.com/rondolab/MR-PRESSO. The default parameters were used for the MR-PRESSO analysis.

In the pQTL summary data, SNPs with a P-value less than 1.0 × 10^–5 were selected as potential instrumental variables. The selection criterion was set to 1.0 × 10^–5 for the two pQTL studies because 5.0 × 10^–8 would lead to too few instrumental variables. We clumped SNPs (LD r² < 0.01 within 10,000 kb) based on data from Europeans from the 1000 Genomes project using the “clump_data” function in the R package TwoSampleMR to select independent instrumental variables. The effect allele of each SNP in the BP GWAS and pQTL studies was manually checked for consistency.

For proteins that passed the three MR tests, we applied the CAUSE method to account for horizontal pleiotropic effects by pathways other than those considered in the multivariable approach. CAUSE is a Mendelian randomization accounting for correlated and uncorrelated horizontal pleiotropic effects using genome-wide summary statistics (37). Data used in the CAUSE analysis were from pQTL studies and GWAS described above. We used 1,000,000 genetic variants to estimate the nuisance parameters. Other parameters were left as their defaults in the CAUSE analysis.

Based on published GWAS papers, we collected up to 162 SCAD-associated genes (Supplementary Table 1). Genes such as FBN1 and PHACTR1 were confirmed in several GWAS. The most recent GWAS published at Nature Communication identified many novel genes and confirmed previously identified genes. In this GWAS, we selected SNPs in 13 SCAD-associated loci (Table 1). The top signals were located in 1q21.2, in which 149 SNPs showed significant associations with SCAD (P < 5.0 × 10^–8). The top SNP associated with SCAD in 1q21.2 was rs12740679 (P = 2.19 × 10^–12). There were fewer genome-wide significant SNPs in other selected loci. Five SNPs in 12q13.3 and one SNP in 6p24.1 were significantly associated with SCAD. For the remaining ten selected loci, no SNPs reached the genome-wide significance level. For 2p23.2, 4q34.3, 5q23.2, and 12q21.33, only one SNP reached 1.0 × 10^–5 in each locus. In 6q25.3, 11 SNPs passed 1.0 × 10^–5. The top SNP in 6q25.3 was rs78349783 (1.03 × 10^–6).

We selected 230 thousand SNPs inside the 162 SCAD-associated genes according to the UCSC database. Then we looked up associations between these SNPs and BP in GWAS summary data. We found 204 and 228 SNPs that were significantly associated with SBP and DBP (P < 5.0 × 10^–8), respectively. These SNPs were located in 11 genes, including JAG1, ERI1, ULK4, THSD4, CMIP, COL4A2, FBN1, FAM76B, FGGY, NUS1, and HNF4G. A total of 111 and 114 SNPs in the ULK4 gene were associated with SBP and DBP (P < 5.0 × 10^–8), respectively. SNPs in JAG1 and ERI1 both associated with SBP and DBP. SNPs in FGGY, HNF4G and NUS1 were associated with SBP; SNPs in CMIP, COL4A2, FAM76B, FBN1, and THSD4 were associated with DBP. These SNPs were located in SCAD-associated genes, and therefore were considered to be shared SNPs for SCAD and BP. We next searched whether these SNPs were associated with circulating proteins.

We looked for pQTL signals in five studies. We found significant signals in three studies (20, 21, 23) for the shared SNPs (Table 2). A total of 49 pQTL signals were identified. The SNPs were located in FAM76B, COL4A2, JAG1, ITSN1, ERI1, and CMIP. The regulated proteins were encoded by 13 genes, including MMP10, IL6R, FIGF, MMP1, CTSB, IGHG1, DSG2, TTC17, RETN, POMC, SCARF2, RELT, and GALNT16. BP-associated SNPs in ERI1 were strongly associated with circulating cathepsin B (CTSB), and evidence was found in KORA and INTERVAL studies (21, 23).

To examine the potential biological functions of the proteins affected by SCAD and BP-associated SNPs, we performed GO analysis in the DAVID database. Among the 11 shared genes, COL4A2, THSD4, and FBN1 were related to extracellular matrix components (P = 1.10 × 10^–3) and proteinaceous extracellular matrix (P = 8.60 × 10^–7). The 13 BP- and SCAD-associated proteins were enriched in specific GO biological process, cellular component and molecular function terms (false discovery rate < 0.1), including cellular components such as extracellular space (P = 2.70 × 10^–3); biological process such as collagen metabolic process (P = 5.70 × 10^–7), multicellular organism metabolic process (P = 6.90 × 10^–7); molecular function such as serine-type peptidase activity (P = 6.50 × 10^–4), serine hydrolase activity (P = 6.70 × 10^–4) and endopeptidase activity (P = 3.30 × 10^–3) (Table 3). Proteins encoded by IGHG1, MMP1, MMP10, CTSB, POMC, RETN, and IL6R may have functional roles in BP regulation and the development of SCAD.

We tested whether the seven proteins were genetically associated with BP using several MR methods (Table 4). We found that the association between circulating cathepsin B level and DBP was significant in MR-Egger analyses using data from KORA and INTERVAL studies. Significant association between circulating cathepsin B level and DBP was also found in the weighted median analysis using data from INTERVAL study. After adjusting for outliers in MR-PRESSO analysis, the associations between circulating cathepsin B level and DBP was marginal significant (P = 0.0580). Significant association between circulating cathepsin B level and SBP was found in the weighted median analysis using data from INTERVAL study. Furthermore, we corrected for correlated and uncorrelated horizontal pleiotropy using the CAUSE method and retained an indication for the causal effect of cathepsin B on BP (Figure 2). By using INTERVAL data, we found that the causal model was significantly better than the null and sharing models for SBP (P = 0.020 and 0.024, respectively). The instrumental variables used in the MR analyses were presented in Supplementary Table 2.

For aortic remodeling, we first examined the association between BP and AAD and aortic diameters by MR analysis (Table 5). Significant associations between DBP and AAD, AA and DA diameters were found. For SBP, significant association between SBP and AA diameter were found in weighted median analysis and MR-PRESSO analysis. Marginal significant association between DBP and AAD was found in the CAUSE analysis (P = 0.049). Significant association between DBP and AA diameter was found in the CAUSE analysis (P = 7.48 × 10^–4) (Figure 3). The instrumental variables used in the MR analyses were presented in Supplementary Table 3.

We then examined the association between circulating cathepsin B level and AAD and aortic diameters by MR analysis (Table 4). Significant association between circulating cathepsin B level and AAD was not found. We found that significant associations between circulating cathepsin B level and AA diameter in weighted median (P = 8.34 × 10^–5), IVW (P = 3.63 × 10^–3) and MR-PRESSO (P = 9.38 × 10^–3) analyses by using the INTERVAL data. We found that nominal associations between circulating cathepsin B level and AA (P = 0.038) and DA (P = 0.067) diameters in the CAUSE analysis by using the INTERVAL data. The instrumental variables used in the MR analyses were presented in Supplementary Table 4.